Section
Practice reports from systems we run — what we changed, what it cost, and what we measured afterwards.
6 articles
DevOps9 min read
Traditional monitoring answers whether a request succeeded. A feature that returns a confident, well-formed, incorrect answer succeeds every time — so the signals have to be different.
Software Architecture9 min read
We measured what a repository layer, a service layer and a DTO mapper actually cost across 14 months of changes. The answer was 38% of every ticket.
Software Architecture10 min read
A schema change is easy. A schema change deployed across a fleet where old and new code run simultaneously, without a maintenance window, is a different problem with a known shape.
Software Architecture9 min read
Flags are cheap to add and nobody owns removing them. A codebase with three hundred of them has 2^300 configurations, none of which is tested, and one of which is production.
Software Engineering8 min read
A test that has never failed is either protecting something or costing something. We spent three weeks finding out which, and deleted 2,140 of them.
DevOps10 min read
Three retries at each of four layers is 81 requests for one user action. We traced an outage back to our own clients and rewrote the policy.